Investigating Map-Based Path Loss Models: A Study of Feature Representations in Convolutional Neural Networks

TL;DR

This paper explores how different feature representations in convolutional neural networks affect path loss prediction accuracy using map-based models, finding that image channel encoding of scalar features enhances generalization.

Contribution

It systematically compares scalar feature encoding methods in CNNs for path loss modeling, highlighting the superiority of image channel representation.

Findings

Image channel encoding improves model generalization.

Scalar features as image channels outperform scalar inputs.

Different feature configurations significantly impact prediction accuracy.

Abstract

Path loss prediction is a beneficial tool for efficient use of the radio frequency spectrum. Building on prior research on high-resolution map-based path loss models, this paper studies convolutional neural network input representations in more detail. We investigate different methods of representing scalar features in convolutional neural networks. Specifically, we compare using frequency and distance as input channels to convolutional layers or as scalar inputs to regression layers. We assess model performance using three different feature configurations and find that representing scalar features as image channels results in the strongest generalization.